GEO

AI search optimization: the 2026 guide

How to get cited by ChatGPT, Gemini, Perplexity, Claude, Copilot and Google's AI features: crawler access, citable pages, third-party proof and measurement.

By Ibrahem HelalAI-assisted, reviewed by Jeremy ZavaPublished 23 September 202616 min read

AI search optimization is the work of getting your pages and your brand cited when someone asks ChatGPT, Gemini, Perplexity, Claude, Copilot or Google's AI features instead of scanning ten blue links. It needs three things: every AI crawler let in, pages an engine can lift a complete answer from, and third-party pages that already vouch for you.

That is the method behind our GEO services. This guide takes each engine apart, covering where it gets its answers, which crawler it sends and what it rewards, and then covers the work that is the same everywhere: crawler access, page format, third-party proof, measurement, cost and timeline.

On this page

What is AI search optimization, and how is it different from SEO?

AI search optimization is SEO with a different finish line. Classic SEO ends when your page ranks in a list; AI search optimization ends when an assistant quotes your page, names your brand, or links you as a source inside a generated answer. The foundations are shared, and the last mile is not.

The shared part is large. Every engine in this guide needs a page it can crawl, read without running scripts, and index. Google is explicit that its AI features run on ordinary search eligibility and nothing more [1]. If your technical SEO is broken, no amount of AI-specific work fixes it.

The different part is where the work goes after that. A search engine ranks whole pages; an assistant lifts passages, usually two or three sentences, and assembles them with passages from other sites. It also reads a wider set of sources than a ranking does, so a review platform or a professional profile can be doing more for you than your own homepage. And it is not deterministic: the same question asked twice returns different sources, so you manage a rate rather than a position.

You will see the same work under other labels. Generative engine optimization, answer engine optimization and LLM optimization all describe roughly this, and GEO vs SEO vs AEO sets out where the terms diverge. Our generative engine optimization playbook is the strategy-level version; this guide is the engine-by-engine version, and AI search ranking factors lists the signals in one place.

One thing to settle before you spend money: which engines your buyers use. Ask ten customers what they typed and where. The answer decides whether you start with Google's AI features, which most buyers still meet first, or with ChatGPT, which is where most of the noise is. A market-share chart is a poor substitute for that conversation.

Which AI engines matter, and where does each one get its answers?

Seven surfaces matter for most businesses: ChatGPT, Google AI Overviews, Google AI Mode, the Gemini app, Perplexity, Claude and Microsoft Copilot. Each one retrieves from a different place, and that is the fact the whole discipline rests on. There is no single AI index to optimise for; there are several, built by different crawlers, with different habits.

Google's surfaces share one index. AI Overviews and AI Mode pull from Google Search using a query fan-out technique, issuing several related searches across subtopics to build one response [1]. The Gemini app grounds its answers in Google Search too, but its training and grounding are governed by a separate robots.txt token, Google-Extended, which has no effect on Search itself [4].

OpenAI runs its own index. OAI-SearchBot builds the index that ChatGPT search cites from, ChatGPT-User fetches a page when a user asks about it, and GPTBot collects training data; the three are controlled independently [3]. Anthropic split its crawling the same way: Claude-SearchBot indexes for search, Claude-User fetches on a user's behalf, and ClaudeBot gathers training data [5].

Perplexity runs PerplexityBot for its own index and Perplexity-User for live visits, and says neither is used to train foundation models. Copilot answers from Bing's index, so Bingbot is the crawler that matters there.

The engines also differ in how much they cite. In a study tracking 13,184 citations across four assistants, Perplexity averaged 19.2 sources per answer, Gemini 8.1, ChatGPT 4.5 and Claude 3.6 [2]. Competing for one of nineteen slots is a different job from competing for one of four.

Read each engine's habits in its own post. How Perplexity ranks sources, how Claude chooses sources and how AI Overviews choose sources are the three explainers, and ChatGPT search vs Google covers whether the two are converging or pulling apart.

Do the AI engines cite the same sources?

Almost never. Across 1,792 question-and-domain combinations in the tracking study, all four major assistants agreed on citing the same domain for the same question thirty times, which is 1.7% [2]. Even ChatGPT and Perplexity, the two most used, shared between 5% and 8% of their cited domains on the brands tested [2].

30 in 1,792

They do not even agree with themselves. The same seventy-two questions run two days apart produced different sources from every engine: Claude kept only 30% of the URLs it had cited, Gemini kept 38% of its domains, and Perplexity, the most stable, kept 65% [2]. A screenshot of one answer is not evidence of anything.

Two consequences follow. First, you plan per engine rather than for AI in general, because the page that wins on Perplexity may be invisible to ChatGPT. Second, you measure a rate across many prompts and many runs, never a single result, which is why the measurement section below insists on a fixed prompt set.

The pairwise differences are worth knowing in detail if two engines dominate your market. ChatGPT vs Perplexity citations compares the two engines that cite most differently, and Gemini vs ChatGPT for brand visibility covers the pair most buyers actually switch between.

Can the AI crawlers reach your site?

Check this before anything else, because a blocked crawler makes every other measurement in this guide meaningless. Most sites that are invisible in AI answers are not losing on content; a firewall rule, a hosting toggle or an over-broad robots.txt line is quietly refusing the bots.

The mistake is easy to make because the crawlers are not what people assume. Each vendor now runs separate bots for training, for building a search index and for fetching a page during a conversation, and each is controlled by its own robots.txt token [3][5]. A rule written in 2024 to keep GPTBot out of your training data also keeps you out of ChatGPT's answers if it was written as a wildcard. The full list, with what each bot does, is in the AI crawlers list.

Then there is the edge. Cloudflare, and several hosts, ship a one-click control that blocks every AI user agent at once, retrieval crawlers included. It produces no error and no visible traffic drop. Cloudflare AI bot blocking walks through the setting and how to allow the crawlers you want while keeping the scrapers out.

One token deserves its own paragraph because it is so widely misread. Google-Extended controls whether Google may use your content to train Gemini models and to ground answers in the Gemini app and Vertex AI; it has no effect on Google Search, AI Overviews or AI Mode [4]. Blocking it does not protect you from AI Overviews, and allowing it does not get you into them. Google-Extended explained covers what it does and does not do.

What does a page need before an AI engine will cite it?

A page needs one passage that answers one question completely in two or three sentences, and it needs to be easy to find, easy to lift and easy to trust. Engines quote passages, not pages, so the unit you are editing is the first few sentences under each heading.

That reframes the usual content advice. Length is not a lever; a longer page has more candidate passages, not better ones. Keyword density is irrelevant.

What matters is whether a claim can stand alone when pasted into a reply without the paragraph above it. If a sentence needs context to make sense, it will not be lifted, or it will be lifted wrongly.

Structure does the rest. Phrase headings as the questions buyers type, because those are also the sub-questions Google's fan-out generates [1]. Open each section with the answer, then the detail.

Put a table on the page, because a comparison is the easiest thing to extract and the hardest to paraphrase badly. Give numbers with a named publisher and a date beside them, and sign the page with a real person who has a page of their own.

The extraction test takes a minute. Read any section aloud from its first sentence and stop after three. If what you said is a correct, complete answer to the heading, the section is quotable. If you were still setting up, nothing further down will rescue it; rewrite the opening.

Pages that hide the useful number lose most often. If the question is what something costs and your page answers with a form, you have handed the citation to whichever competitor printed a figure. That applies to timelines, inclusions, exclusions and minimum terms as much as to price. AI search ranking factors ranks the signals by how much evidence sits behind each one.

How do you show up in Google AI Overviews and AI Mode?

By being indexed, eligible for a snippet, and the best answer to one of the sub-questions Google generates while building the response. Google's documentation says there are no additional requirements to appear in AI Overviews or AI Mode and no special optimizations are necessary [1]. That is the whole technical bar, and it is worth believing.

What changes is the question being answered. Both features may use query fan-out, issuing multiple related searches across subtopics and data sources, and identifying more supporting pages while the response is being generated [1]. So a page that answers one narrow sub-question well can be pulled into an answer about something broader, and a page that covers the head term loosely can be skipped for one that nails a fragment of it.

The practical method is to write the question tree. Take the commercial question your buyer has and list the eight or ten sub-questions they would need answered to decide: price, timeline, what is included, what is not, alternatives, risks, who else uses it. Each becomes a heading, and each heading's first sentences become a candidate passage. Rank in Google AI Overviews applies that to Overviews, Google AI Mode optimization covers the conversational surface where fan-out runs deepest, and how AI Overviews choose sources explains why a page that ranks fourth is sometimes cited and a page that ranks first is not.

Because these surfaces run on Google's index, they are also where an ordinary SEO problem shows first. A page dropped from the index, a canonical pointing the wrong way, a slow template that stopped being crawled: each removes you from AI answers before it shows in rankings. Our AI Overviews and AI Mode service starts with that plumbing for the same reason.

How do you get cited by ChatGPT, Perplexity, Claude, Gemini and Copilot?

Give each engine the kind of source it already reaches for. The habits differ, and they are consistent enough to plan around: ChatGPT skews to primary sources, Perplexity to professional and community pages, Claude to review sites and curated lists, Gemini to community threads and roundups [2].

ChatGPT cites few sources and prefers documentation, research and vendors' own pricing and help pages [2]. The good news for a company with a real product is that your own pages are eligible; the bad news is that a page written to persuade is not a primary source. How to get cited by ChatGPT is the full method, starting with the three OpenAI crawlers.

Perplexity cites the most and reads the widest. Its most-cited domains in the tracking data were LinkedIn, YouTube and Reddit [2], which means a complete company profile and a video that answers the buyer's question can earn a citation your website cannot. How Perplexity ranks sources covers what its index rewards.

Claude cites the fewest sources and leans on review platforms and curated listicles; in the tracking data the review platform Clutch was its single most-cited third-party domain [2]. If your category is reviewed anywhere, that profile is your Claude strategy. How Claude chooses sources has the detail.

Gemini leans on community posts and roundup lists [2], and because the app grounds in Google Search, ordinary Google visibility carries over. Allowing Google-Extended matters here in a way it does not for AI Overviews [4]. Get cited by Gemini covers the app specifically.

Copilot answers from Bing's index, which most sites never look at. Register in Bing Webmaster Tools, submit through IndexNow, and check Bing's own AI performance reporting for what Copilot is citing. It is cheap insurance rather than a strategy, and get cited by Microsoft Copilot sets out what the Bing side needs.

Do not try to do all six at once. Pick the engine your buyers named, fix its crawler and its favourite source type first, and let the shared foundations carry the rest.

What about products, shopping and local recommendations?

Product and local questions are answered from data feeds and listings more than from web pages, so the work moves off your blog and into your catalogue and your business profile. An assistant asked for the best running shoe under $150 or a plumber nearby is not reading your homepage; it is reading structured product data, merchant listings, map profiles and reviews.

For products, ChatGPT's shopping answers draw on merchant-provided product data alongside what it finds on the web, so the feed matters: accurate titles, prices, availability, images and identifiers, kept in agreement with the product pages. A page that says one price and a feed that says another is a product the engine will not recommend with confidence. ChatGPT shopping visibility covers the merchant side, what the feed needs and how product pages should be written to be lifted.

For local businesses, the inputs are the profile, its category, its hours and its reviews. Assistants answering near-me questions lean on the same listing data that maps products use, and on review platforms for the recommendation itself. Consistency does the heavy lifting: one legal name, one address, one phone number, spelled identically everywhere. ChatGPT recommendations for local businesses sets out which profiles matter and what to write in them.

Where do AI citations come from, if not your own site?

From the pages other people write about your category. Across the tracking study, the domains cited most were professional networks, video, community platforms, review sites and curated lists, and on at least one engine a review platform was the single most-cited third-party domain [2]. Your own site competes with all of them for a handful of slots.

That has an awkward implication for the budget: some of the highest-return work is not content work. It is claiming and completing the review profiles in your category, keeping the company and founder profiles current on the professional networks, and being included on the merits in the comparison posts that already rank.

Rank that work by how fast it pays. Profiles you control take an afternoon each and are permanent. Inclusion in other people's lists takes weeks and a low hit rate, so start asking early. Publishing your own comparison, with a stated method and a plain disclosure of where you sit in it, is slowest and is the only one of the three that other people will cite without being asked.

One caution about forums. Community sites are among the most-cited domains overall, yet in the tracking study they were nearly absent from citations for business-buying questions on two of the four engines [2]. If your buyer is a company rather than a consumer, spend the effort on review platforms and listicles first.

When a brand is missing from AI answers, the fix is rarely another blog post. It is usually three or four third-party pages the engines keep reading that either do not mention the brand or describe it wrongly. Fix those and the citations follow.

How do you know it is working?

By tracking a citation rate and a mention rate, per engine, over a fixed set of questions, and by reading the first-party reports the search engines now publish. A single answer is noise; the same prompts asked two days apart return different sources from every engine [2].

Set the prompt set before you touch anything. Twenty to forty questions a real buyer would ask, run weekly, recording whether you were named, whether you were linked, and which URL was linked. Keep the questions fixed even when the answers are unflattering, because changing them resets the baseline and destroys the trend. The AI brand visibility guide is the long version of that method.

Add the free sources. Google Search Console and Bing Webmaster Tools both report on how their AI surfaces interact with your pages, and that data is about your site rather than about a sample of prompts somebody chose. It will not tell you what an assistant said about you, but it will tell you whether anything is reaching you from those surfaces at all.

Then write one sentence a month explaining each movement, and keep the sentences. Six months of one-line explanations is the only record of what you changed and what happened next. An AI visibility audit is how we take that first baseline before any content work starts.

What does AI search optimization cost, and how long does it take?

Expect the crawler and access fixes to pay in weeks, page rewrites in one to two months, and third-party presence in a quarter or more, because that part depends on other people. Costs follow the same shape: the first phase is a few hours of technical time, the second is writing, and the third is patient outreach.

Bought as a retainer, the prices you will see range from a few hundred dollars a month for a tracking tool with no labour attached, through low four figures for a programme run alongside SEO, to five figures where the scope includes original research and digital PR. The wide range is mostly scope, not quality, and the honest way to compare quotes is to ask for the scope in units: which crawlers were checked, which pages rewritten, which profiles claimed, how many prompts tracked.

We sell this work, so read the next sentence knowing that. Our plans carry their starting prices on the pricing page, and we print them for the same reason we told you to print yours: a number on a page can be quoted by an engine and argued with by a buyer, and a number behind a form can do neither.

Three things to resist along the way. Do not buy a second tracking tool because the first showed a bad week. Do not publish volume on the theory that more pages mean more chances, because the engines count passages, not pages, and Google's quality bar applies unchanged to its AI features [1]. And do not rewrite the prompt set because the answers stung.

What this guide covers next

The rest of the cluster takes each engine, each control and each comparison in turn, and every post below links back here.

The how-to guides first. Get cited by Gemini covers the app that grounds in Google Search but answers differently from it. Get cited by Microsoft Copilot covers the Bing side most sites have never set up.

Rank in Google AI Overviews applies the fan-out method to the surface most buyers see first, and Google AI Mode optimization covers the conversational surface where follow-up questions decide who gets cited. ChatGPT shopping visibility is for merchants with a catalogue, ChatGPT recommendations for local businesses is for anyone who depends on near-me answers, and Cloudflare AI bot blocking shows how to keep the scrapers out without locking the assistants out too.

Then the explainers. How AI Overviews choose sources, how Perplexity ranks sources and how Claude chooses sources each describe one engine's selection habits in under a thousand words, and Google-Extended explained settles what that token does and does not control.

Then the comparisons. ChatGPT vs Perplexity citations compares the engines that cite most differently, Gemini vs ChatGPT for brand visibility compares the two assistants most buyers switch between, and ChatGPT search vs Google asks whether one is replacing the other and what the answer means for a marketing budget.

Finally the reference lists. AI search ranking factors ranks the signals by the evidence behind them, and the AI crawlers list names every bot, what it does and the robots.txt token that controls it.

Start with the crawler check, whichever engine you care about most. Everything else in this cluster assumes the bots can get in.

Questions people also ask

• Is AI search optimization the same as GEO?

  In substance, yes. Generative engine optimization, answer engine optimization and AI search optimization all describe getting cited inside generated answers. What this guide adds to the generic playbooks is the engine-by-engine detail: where each one retrieves from, which crawler it sends and what it rewards.

• Which AI crawlers should I allow in robots.txt?

  All of the retrieval and live-fetch bots: OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, Claude-SearchBot, Claude-User, Googlebot and Bingbot. Decide separately on the training bots, GPTBot, ClaudeBot and Google-Extended, because those govern what the models learn rather than what they cite.

• Does blocking Google-Extended keep me out of AI Overviews?

  No. Google-Extended controls training and grounding for the Gemini app and Vertex AI only. AI Overviews and AI Mode use Googlebot and the ordinary Search index, so the only way to stay out of them is the usual snippet and noindex controls.

• How long before AI search optimization shows results?

  Crawler fixes show within weeks once pages are recrawled. Page rewrites take one to two months to be picked up. Third-party presence, which drives most brand mentions, takes a quarter or more because it depends on other people publishing.

• Do I need to rank in Google to be cited by AI engines?

  Not for most of them. In the tracking study this guide cites, only a minority of the domains the assistants cited also appeared in Google's top ten for the matching query. Google's own AI features are the exception, because they run on Google's index.

• Can a tool do this for me?

  A tool tells you where you stand across engines. It does not let the crawlers in, rewrite the page or earn the third-party mention, which is where the work is. Budget for it as measurement, not as the programme.

Sources

1. AI features and your website — Google Search Central (checked 22 September 2026)
2. I tracked 13,184 AI citations across 4 LLMs: here's what I learned — Orbit Media Studios (checked 22 September 2026)
3. Overview of OpenAI crawlers — OpenAI (checked 22 September 2026)
4. Google's common crawlers — Google Search Central (checked 22 September 2026)
5. Does Anthropic crawl data from the web, and how can site owners block the crawler? — Anthropic (checked 22 September 2026)

Find out which AI engines can see your site, and which cannot

Book a free 30-minute consultation with Ibrahem and Jeremy. We look at your site while we talk, and you leave with a prioritised plan.

Book a free 30-minute consultation

Who wrote this

Ibrahem HelalCo-founder

Leads strategy and delivery.

Works on: SEO, GEO

Reviewed before publishing by Jeremy Zava, Co-founder.

Posts by Ibrahem Helal

More on this

• AI Search Ranking Factors That MatterAI engines do not use one ranking algorithm, but the pages they all reach for share nine traits: primary, specific, fluent and easy to lift.
• The AI crawlers list: every bot to knowGPTBot, ClaudeBot, PerplexityBot, Google-Extended and more: what each AI crawler does, what blocking it costs you, and how to check your rules.
• ChatGPT Search vs Google: What's DifferentGoogle still handles the overwhelming majority of searches worldwide, but ChatGPT wins a specific kind of question — and the two barely cite the same pages.
• How to Rank in Google AI OverviewsAI Overviews draw on the same index and ranking systems as regular Search results, so the fastest way in is fixing what already keeps you out of the top ten.